How I solved ChatGPT's hallucination problem for K-Beauty ingredients (and turned it into an API)

Hi everyone,

I’ve been working with Korean cosmetics data and realized a huge problem: general AI models (like ChatGPT) hallucinate badly when translating Korean cosmetics ingredients into global INCI standards.

They literally invent fake chemical names for Asian herbal extracts or marketing terms. If you are building a beauty app or handling customs clearance, this is a legal nightmare.

So, I built a bulletproof API to fix this. Here is how I solved the hallucination issue:

1. The Zero-Error Dictionary Fallback: I hardcoded a massive dictionary of verified ingredients. The API checks this first. Zero AI guesswork, pure facts.

2. Strict AI Prompting: For unknown ingredients, it uses a strictly-prompted AI model. If the AI is even 1% unsure, it is forced to refuse to guess and explicitly returns UNVERIFIED_MANUAL_CHECK_REQUIRED.

Now, it only returns 100% verified data and clearly flags what needs a human review.

I just published the MVP on RapidAPI. It has a freemium tier (100 free calls/month) so you can test it out without a credit card.

Link: https://rapidapi.com/dahee8703dahee8703/api/legal-k-beauty-inci-translator

I would love to hear your feedback! Has anyone else struggled with domain-specific AI hallucinations? Let me know what you think.